Thomas Hoffmann wrote:

>On 29 Jun 2005 at 10:44, Joseph Hilferty wrote:
>
>  
>
>>Whoa! Not so fast here! You can do this by linking directly to
>>the verb's valence. No need for traces.
>>    
>>
>
>As my presentation will hopefully show, I'm aware of the analysis 
>that you can dispense with traces if you assume something like SLASH-
>features on the V-head. As far as I know, the data supporting such an 
>analysis exclusively comes from processing studies (Pickering et 
>al.). So I was wondering whether production data might also support 
>this view or whether you get contradictory evidence.
>  
>
If I recall correctly, the Traxler and Pickering paper (not the Barry 
and Pickering paper)
uses as stimuli certain examples that are adjuncts and not really 
complements (or at least
in the fuzzy boundary between the two).

>>In any event, an example like Thomas's:
>>
>>(1) * The place in which I live in
>>
>>superimposes two (conflicting) structures:
>>
>>   The place              I live  in
>>   The place   in which   I live
>>    
>>
>
>That's actually an interesting question: does the syntactic 
>"generator" assemble more than one syntactic structure?
>
Probably. Semantic priming studies give good reason to believe this.
I think that it is a safe assumption that spreading activation occurs
in syntax also.

> So do we have 
>OT-like competition between these structures (and possibly "which I 
>live in", and "that I live in" ... as well)? Then in the the doubled 
>P-cases, the processor would obviously conflate two representations 
>into a single one. However, I must confess that I'm a bit seceptical 
>towards OT approaches. I think a generative HPSG-construction grammar 
>approach gets the same results would postulating these vast number of 
>potential competitors. [@Joe: I didn't mean to imply that you were 
>advocating such a position, it was just a thought that came to mind, 
>when I saw you're example] 
>  
>
Construction grammar could do this if the inheritances were weighted
somehow. The same would go for HPSG, I assume. But that would bring
these theories into the realm of performance, and I'm not sure they're
ready for that.

BTW, WG should, in principle, be able to conflate structures through
spreading activation.

Memory has got to be involved with these performance errors somehow.

Joe